588 Diabetes Care Volume 38, April 2015

Carolyn T. Thorpe,1,2 Walid F. Gellad,1,3 Tight Glycemic Control and Use of Chester B. Good,1,2,3,4 Sijian Zhang,1 Xinhua Zhao,1,4 Maria Mor,1,5 and Hypoglycemic Medications in Michael J. Fine1,3 Older Veterans With Type 2 Diabetes and Comorbid Dementia Diabetes Care 2015;38:588–595 | DOI: 10.2337/dc14-0599

OBJECTIVE Older adults with diabetes and dementia are at increased risk for and other adverse events associated with tight glycemic control and are unlikely to experience long-term benefits. We examined risk factors for tight glycemic control in this population and use of medications associated with a high risk of hypoglycemia in the subset with tight control.

EPIDEMIOLOGY/HEALTH SERVICES RESEARCH RESEARCH DESIGN AND METHODS This retrospective cohort study of national Veterans Affairs (VA) administrative/ clinical data and Medicare claims for fiscal years (FYs) 2008–2009 included 15,880 veterans aged ‡65 years with type 2 diabetes and dementia and prescribed antidi- abetic medication. Multivariable regression analyses were used to identify socio-

demographic and clinical predictors of hemoglobin A1c (HbA1c) control (tight, moderate, poor, or not monitored) and, in patients with tight control, subsequent use of medication associated with a high risk of hypoglycemia (sulfonylureas, insulin).

RESULTS 1Center for Health Equity Research and Promo- Fifty-two percent of patients had tight glycemic control (HbA1c <7% [53 mmol/mol]). fi tion, Veterans Affairs Pittsburgh Healthcare Sys- Speci c comorbidities, older age, and recent weight loss were associated with tem, Pittsburgh, PA greater odds of tight versus moderate control, whereas Hispanic ethnicity and obe- 2Department of and Therapeutics, sity were associated with lower odds of tight control. Among tightly controlled University of Pittsburgh School of Pharmacy, Pittsburgh, PA patients, 75% used sulfonylureas and/or insulin, with higher odds in patients who 3 ‡ Division of General Internal , University were male, black, or aged 75 years; had a hospital or nursing home stay in FY2008; of Pittsburgh School of Medicine, Pittsburgh, PA or had congestive heart failure, renal failure, or peripheral vascular disease. 4Veterans Affairs Pharmacy Benefits Manage- ment, Chicago, IL CONCLUSIONS 5Department of Biostatistics, University of Pitts- Many older veterans with diabetes and dementia are at high risk for hypoglycemia burgh Graduate School of , Pitts- burgh, PA associated with intense diabetes treatment and may be candidates for deintensi- fi Corresponding author: Carolyn T. Thorpe, ctthorpe@ cation or alteration of diabetes medications. pitt.edu. Received 7 March 2014 and accepted 7 Decem- The prevalence of diabetes among adults aged $65 years is projected to increase ber 2014. dramatically by 2050 (1). Dementia affects up to 16% of diabetic patients aged $65 This article contains Supplementary Data online years and 24% aged $75 years (2,3), and evidence shows that the two conditions at http://care.diabetesjournals.org/lookup/ suppl/doi:10.2337/dc14-0599/-/DC1. share a pathophysiological link (4–6). As described in the federally mandated Na- ’ © 2015 by the American Diabetes Association. tional Plan to Address Alzheimer s Disease (7), U.S. health care is poorly situated to Readers may use this article as long as the work address the needs of older adults with dementia and common comorbidities such as is properly cited, the use is educational and not diabetes. This is evident in the two- to threefold increased odds of severe for profit, and the work is not altered. care.diabetesjournals.org Thorpe and Associates 589

hypoglycemia and preventable hospital- chlorpropamide in all patients aged FY2008 (1 October 2007) and 2)were izations in diabetic patients with comor- $65 years (20). However, sulfonylureas given two or more inpatient or outpa- bid dementia (2,8). Improving ambulatory and insulin are recommended as first- tient diagnoses for type 2 diabetes (ICD-9 care, particularly reducing known risk fac- and second-line diabetes in 250.x, 250.x2) in FY2008–2009 (with first tors for hypoglycemia, in older patients the general patient population due to diagnosis in FY2008) or received an oral with coexisting diabetes and dementia is strong evidence of efficacy in lowering diabetes medication through the VA in critical, but to date, few such efforts have HbA1c, low cost, market longevity, and FY2008 (25). We then linked to Medicare been made. low risk of adverse events apart from claims and enrollment data to further re- There is growing consensus that be- hypoglycemia (17,19). As patients de- fine the sample (Fig. 1). We applied the cause of their increased risk for hypogly- velop dementia, the risk of hypoglycemia Medicare Chronic Conditions Ware- cemia and reduced potential to benefit increases and the risk-to-benefit balance house ICD-9 diagnosis code list for from tight glycemic control, older diabetic of using these agents changes, especially Alzheimer’s Disease and Related Disorders patients with dementia should avoid tight if HbA1c is tightly controlled. To our (ADRD) to both VA records and Medicare glycemic control (HbA1c ,7% [53 mmol/ knowledge, no prior studies have exam- claims to identify patients with dementia mol]) and instead pursue moderate HbA1c ined the prevalence of or risk factors for (26,27). The ICD-9 codes in this algorithm levels of 7% to ,9% (53 to ,75 mmol/ the use of medications with a high risk of include Alzheimer disease, vascular de- mol). Older individuals with dementia of- hypoglycemia in patients with dementia. mentia, and a range of other specific re- ten have other risk factors for hypoglyce- In this study, we addressed these lated disorders (see Supplementary mia, including weight loss, changes in gaps in the literature by identifying risk Table 1 for full list of diagnoses). This appetite and eating habits, and difficulty factors for tight glycemic control in definition contains only minor differen- following prescribed regimens (9–11), older veterans with dementia receiving ces from an algorithm shown to have and a reduced ability to recognize and antidiabetic medication . We good sensitivity and specificity com- respond appropriately to symptoms also identified the prevalence and char- pared with a gold standard clinical de- (9,12,13). Furthermore, older patients acteristics of patients at highest poten- mentia assessment (27). Patients with dementia have an average life ex- tial risk for hypoglycemia through their without an ADRD ICD-9 code who pectancy of 2–8 years (14–16) and are use of sulfonylurea and/or insulin after filled a VA prescription for an antide- unlikely to experience benefits of tight exhibiting tight glycemic control. Such mentia medication (galantamine, riva- glycemic control, which take years to ac- data can help to inform future interven- stigmine, donepezil, memantine, or crue and are less likely in patients with tions to improve adherence to the VA tacrine) in FY2008 were also included. long-standing diabetes (11). As such, diabetes treatment guidelines and AGS Next, because veterans aged $65 years since 2003, guidelines from the U.S. Choosing Wisely guidelines recommend- are eligible to enroll in Medicare and use Departments of Veterans Affairs and De- ing against tight glycemic control in this non-VA health care in addition to VA care fense (VA/DoD) have presented a risk- population and enhance safer prescribing (yet we did not have access to non-VA stratified approach to glycemic control, for veterans with dementia previously prescription drug records), we took two recommending tight control (HbA1c shown to be at especially high risk for steps to restrict the sample to patients ,7% [53 mmol/mol]) only for patients severe hypoglycemic events (2). with diabetes managed primarily within with a life expectancy of 10–15 years or the VA: 1) eliminating all patients whose more and absent/mild microvascular RESEARCH DESIGN AND METHODS Medicare records indicated enrollment complications (17,18). The American Di- Design and Data Sources in a non-VA source of drug coverage dur- abetes Association and American Geriat- We conducted a longitudinal, retrospec- ing follow-up (i.e., Medicare Part D rics Society (AGS) subsequently adopted tive cohort study using administrative stand-alone plan, Medicare Advantage similar recommendations (19,20), with data from the VA health care system and plan, employer-sponsored plan eligible the AGS in 2013 including the avoidance Centers for Medicare & Medicaid Services. for a Centers for Medicare & Medicaid of tight glycemic control in older adults Data sources included were VA Medical Services retiree drug subsidy) and 2) ex- with comorbidities in its Choosing Wisely SAS Datasets, records of dispensed outpa- cluding all remaining patients with HbA1c campaign (21). Although others have tient prescriptions, and laboratory values levels solely monitored outside the VA, documented the elevated risk of hypo- linked to Medicare Part A, Part B, and en- indicated by no HbA1c values in VA re- glycemia in older diabetic patients with rollment data obtained for a larger study cords and at least one procedure code dementia (2), no studies to our knowl- for an HbA test in Medicare claims. (23,24). All baseline variables and HbA1c 1c edge have focused on understanding risk levels were based on fiscal year (FY) 2008 Next, to ensure accurate capture of out- factors for tight glycemic control in pa- data; the first 120 days of FY2009 served as patient medications used in the 120-day tients with comorbid dementia. the follow-up period for determining anti- follow-up period, we excluded patients Choice of antidiabetic medication also diabetic medication use. The VA Pittsburgh who died or resided in a VA or Medicare- may exacerbate risks to type 2 diabetic Healthcare System Institutional Review covered hospital or nursing home for patients with dementia, especially when Boardapprovedthisstudy. $30 days during the follow-up period HbA1c is tightly controlled (22). Sulfonyl- and, as a result, may not have needed ureas and insulin are associated with a Sample to refill an outpatient VA prescription high risk of hypoglycemia (19), and the To construct the sample, we first used during the follow-up period. Finally, 2013 AGS guidelines explicitly recom- VA data to identify all veterans who 1) we excluded all remaining patients mend against the use of glyburide and were age $65 years as of the start of who did not fill at least one VA 590 Intense Diabetes Treatment in Dementia Patients Diabetes Care Volume 38, April 2015

sample as well as recent weight loss (3% of patients) because of an a priori hypothesis that weight loss is an im- portant contributor to tight glycemic control. Detailed information on the specific ICD-9-CM diagnosis codes used to define each comorbid condition can be found at www.hcup-us.ahrq.gov/toolssoft- ware/comorbidity/comorbidity.jsp. We used VA and Medicare records to deter- mine whether patients had at least one in- patient hospital or nursing home stay in FY2008. Finally, we determined whether dementia was documented within the VA, as indicated by either a VA ADRD diagnosis code or a VA prescription fill for an antide- mentia medication versus documented only in Medicare claims. Analytic Approach We examined descriptive statistics for all study variables in the full sample and each glycemic control group. To deter- mine the association of sociodemographic, clinical, and health care utilization factors Figure 1—Sample construction for older veterans with diabetes and comorbid dementia. VHA, with HbA1c control, we used multinomial Veterans Health Administration. logistic regression with robust SEs. Moder- ate control, which reflects guideline-concor- dant care, was the reference category. prescription for antidiabetic medication purposes, we also captured other specific Finally, for patients whose last HbA1c during follow-up. antidiabetic medication classes (bigua- value in FY2008 indicated tight control nides [metformin], thiazolidinediones (n = 8,276), we estimated a logistic re- Measures [TZDs], a-glucosidase inhibitors, megliti- gression model for use of medication Outcomes nides, dipeptidyl peptidase-4 [DPP-4] in- with high hypoglycemic risk. We also con- We extracted the last available HbA1c hibitors, amylin analogs, and GLP-1 ducted a sensitivity analysis using multino- value in FY2008 from VA laboratory data receptor agonists) that patients used in mial logistic regression to model use of ’ to quantify patients glycemic control as this time period and created a summary sulfonylurea but no insulin, use of insulin , tightly controlled ( 7% [53 mmol/mol]), variable for their overall antidiabetic reg- but no sulfonylurea, and use of both sulfo- , moderately controlled (7% to 9% [53 to imen (noninsulin monotherapy, noninsu- nylurea and insulin as separate outcomes, , 75 mmol/mol]), poorly controlled lin multitherapy, insulin alone, or insulin relative to use of neither agent, and report $ $ ( 9% [ 75 mmol/mol]), and not moni- plus other noninsulin agent). the results in Supplementary Table 2. tored (no HbA1c tests recorded in either VA or Medicare data). We also captured Covariate Risk Factors RESULTS whether the HbA1c value was obtained in We assessed sociodemographic, clinical, Sample Characteristics an outpatient versus inpatient setting for and health care utilization factors in re- Table 1 shows characteristics of 15,880 use in sensitivity analyses. lation to tight glycemic control and use of community-dwelling veterans aged $65 We used outpatient VA drug-dispensing high-risk medications. Patient sex, age years with type 2 diabetes and demen- records for the first 120 days of FY2009 to (65–74, 75–84, or $85 years), and race/ tia. Almost all patients were male (99%), identify patient use of antidiabetic medi- ethnicity (non-Hispanic white, non- and 80% were non-Hispanic white. Co- cation associated with a high risk of hypo- Hispanic black, Hispanic, or other) were morbidities were common, including hy- glycemia, defined as having at least one fill determined from VA utilization files. pertension (81%), deficiency anemia for a sulfonylurea and/or insulin. We Missing VA race/ethnicity values were (21%), chronic lung disease (19%), and used a 120-day window to determine the supplemented with the Research Trian- peripheral vascular disease (17%). period prevalence of use of these drugs in gle Institute race code in the Medicare The majority (52%; n = 8,276) of patients closerelationtothelastHbA1c measure- enrollment file (28). Whether patients had tight glycemic control, 36% had mod- ment. A 120-day window was chosen be- had a copay for VA medications in the erate control, 7% had poor control, and 5% cause even individuals obtaining 90-day follow-up period was also extracted. We did not have HbA1c monitored in FY2008. antidiabetic prescriptions would be ex- used the Elixhauser comorbidity mea- Within the tight control group, the mean pected to fill at least one prescription sure (29,30) applied to diagnoses in VA and median HbA1c value was 6.3% (45 during this period, allowing for some po- and Medicare files to identify comorbid- mmol/mol; minimum 3.8% [18 mmol/mol], tential nonadherence. For descriptive ities present in $5% of the present maximum 6.9% [52 mmol/mol], care.diabetesjournals.org Thorpe and Associates 591

Table 1—Characteristics of community-dwelling older veterans with diabetes and comorbid dementia by level of glycemic control All patients Tightly controlled Moderately controlled Poorly controlled Not monitored Patient characteristics (n = 15,880) (n = 8,276) (n =5,669) (n = 1,131) (n =804) Male sex 15,643 (99) 8,157 (99) 5,581 (98) 1,115 (99) 790 (98) Race/ethnicity Hispanic 1,242 (8) 566 (7) 475 (8) 137 (12) 64 (8) White, non-Hispanic 12,629 (80) 6,742 (81) 4,489 (79) 776 (69) 622 (77) Black, non-Hispanic 1,618 (10) 781 (9) 573 (10) 177 (16) 87 (11) Other 391 (2) 187 (2) 132 (2) 41 (4) 31 (4) Age 65–74 years 3,857 (24) 1,882 (23) 1,442 (25) 367 (32) 166 (21) 75–84 years 8,745 (55) 4,657 (56) 3,058 (54) 586 (52) 444 (55) $85 years 3,278 (21) 1,737 (21) 1,169 (21) 178 (16) 194 (24) Has medication copay 9,515 (60) 4,990 (60) 3,431 (61) 638 (56) 456 (57) Comorbidities Congestive heart failure 2,860 (18) 1,416 (17) 1,080 (19) 226 (20) 138 (17) Heart valve disease 1,168 (7) 648 (8) 394 (7) 69 (6) 57 (7) Peripheral vascular disease 2,624 (17) 1,412 (17) 943 (17) 166 (15) 103 (13) Hypertension 12,815 (81) 6,727 (81) 4,646 (82) 945 (84) 497 (62) Chronic lung disease 2,979 (19) 1,600 (19) 1,032 (18) 212 (19) 135 (17) Hypothyroidism 1,628 (10) 884 (11) 588 (10) 109 (10) 47 (6) Renal failure 2,511 (16) 1,266 (15) 941 (17) 211 (19) 93 (12) Solid tumor without metastasis 2,073 (13) 1,110 (13) 755 (13) 137 (12) 71 (9) Obesity 1,080 (7) 517 (6) 432 (8) 102 (9) 29 (4) Weight loss 436 (3) 260 (3) 129 (2) 20 (2) 27 (3) Fluid and electrolyte disorder 2,016 (13) 1,043 (13) 722 (13) 166 (15) 85 (11) Deficiency anemia 3,308 (21) 1,821 (22) 1,168 (21) 203 (18) 116 (14) Psychoses 2,065 (13) 1,105 (13) 709 (13) 166 (15) 85 (11) Depression 2,463 (16) 1,317 (16) 858 (15) 189 (17) 99 (12) Inpatient stay in FY2008 2,416 (15) 1,246 (15) 837 (15) 243 (21) 90 (11) VA documentation of dementia 11,213 (71) 5,869 (71) 3,958 (70) 818 (72) 578 (72)

Last HbA1c value in baseline year (FY2008)§

HbA1c (%) 6.8 (6.3–7.6) 6.3 (6.0–6.6) 7.5 (7.2–8.0) 9.8 (9.3–10.7) N/A HbA1c (mmol/mol) 51 (45–60) 45 (42–49) 58 (55–64) 84 (78–93) N/A Medication use in follow-up period (first 120 days of FY2009) Medication regimen Noninsulin monotherapy 7,298 (46) 4,942 (60) 1,756 (31) 156 (14) 444 (55) Noninsulin multitherapy 3,081 (19) 1,438 (17) 1,337 (24) 180 (16) 126 (16) Insulin alone 3,308 (21) 1,237 (15) 1,502 (27) 421 (37) 148 (18) Insulin plus other agent 2,193 (14) 659 (8) 1,074 (19) 374 (33) 86 (11) Medication class# Insulin 5,501 (35) 1,896 (23) 2,576 (45) 795 (70) 234 (29) Sulfonylurea 8,927 (56) 4,690 (57) 3,204 (57) 548 (49) 485 (60) Metformin 6,487 (41) 3,593 (43) 2,238 (39) 382 (34) 274 (34) TZDs 826 (5.2) 339 (4) 375 (7) 74 (7) 38 (5) a-Glucosidase inhibitors 233 (1) 81 (1) 116 (2) 28 (3) 8 (1) Use of medications with high hypoglycemic risk No insulin/no sulfonylurea 2,842 (18) 2,063 (25) 598 (11) 42 (4) 139 (17) No insulin/yes sulfonylurea 7,537 (47) 4,317 (52) 2,495 (44) 294 (26) 431 (54) Yes insulin/no sulfonylurea 4,111 (26) 1,523 (18) 1,867 (33) 541 (48) 180 (22) Yes insulin/yes sulfonylurea 1,390 (9) 373 (5) 709 (13) 254 (22) 54 (7)

Data are n (%) or median (interquartile range). Tightly controlled, HbA1c ,7% (53 mmol/mol); moderately controlled, HbA1c 7to,9% (53 to ,75 mmol/mol); poorly controlled, HbA1c $9% ($75 mmol/mol); not monitored, no evidence of having received any FY2008 HbA1c tests in VA or Medicare records. N/A, not applicable. §Presented for the 15,076 patients with HbA1c values in FY2008. #Data not shown for use of meglitinides, DPP-4 inhibitors, amylin analogs, and GLP-1 agonists; ,1% of the total sample used these agents.

interquartile range 6.0–6.6% [42–49 HbA1c values did not substantively af- Table 1 also provides information on mmol/mol]). More than 97% of HbA1c fect the distribution of HbA1c values the medication regimens and specific values were obtained in an outpatient seen in the overall sample or within medication classes used by the sample. setting, and exclusion of inpatient glycemic control categories. Overall, noninsulin monotherapy was 592 Intense Diabetes Treatment in Dementia Patients Diabetes Care Volume 38, April 2015

most common (46%) followed by insulin sulfonylureas or insulin, 29% (n = Racial/ethnic minority status and hav- alone (21%), noninsulin multitherapy 1,811) also took at least one other agent ing an FY2008 inpatient stay were (19%), and insulin plus another noninsu- not associated with a high hypoglycemic associated with higher odds of poor gly- lin agent (14%). Sulfonylureas, metfor- risk (data not shown). cemic control, whereas older age and min, and insulin were the most common deficiency anemia were associated classes used, followed by TZDs and Predictors of Glycemic Control Levels with lower odds. a-glucosidase inhibitors. Meglitinides, The multinomial regression model re- vealed patient age, specific comorbidities, DPP-4 inhibitors, amylin analogs, and Predictors of Use of Medications With and race/ethnicity to be independently High Risk of Hypoglycemia GLP-1 receptor agonists were each associated with having tight versus mod- Among the 8,276 patients with tight used very rarely (,1% of the sample). erate glycemic control (Table 2). Com- control, several significant risk factors Overall, 82% (n = 13,038) of the sam- pared with patients aged 65–74 years, emerged for use of medications with ple used a regimen associated with an those who were 75–84 or $85 years high hypoglycemic risk (Table 3). Male increased risk of hypoglycemia, includ- old had higher odds of HbA1c ,7% (53 sex; black race; age 75–84 or $85 vs. ing 47% (n = 7,537) using sulfonylurea mmol/mol). Heart valve disease, chronic 65–74 years; presence of congestive without insulin, 26% (n =4,111)using lung disease, weight loss, and deficiency heart failure, peripheral vascular dis- insulin without sulfonylurea, and 9% anemia were associated with increased ease, or renal failure; and having had a (n = 1,390) using both agents (Table 1). odds of HbA1c ,7% (53 mmol/mol), hospitalization or nursing home stay in Among tightly controlled patients, 75% whereas congestive heart failure, renal FY2008 were all associated with higher (n = 6,213) used such a regimen, includ- failure, and obesity were associated with odds of use of sulfonylureas and/or insu- ing 52% (n = 4,317) using sulfonylurea, lower odds of HbA1c ,7% (53 mmol/mol). lin versus neither agent. Fluid/electrolyte 18% (n = 1,523) using insulin, and 5% (n = Compared with non-Hispanic white pa- disorder or depression was also associ- 373) using both agents. Of the 6,213 tients, Hispanic patients also had lower ated with lower odds of use of a high-risk tightly controlled patients using either odds of HbA1c ,7% (53 mmol/mol). regimen.

Table 2—Factors independently associated with level of HbA1c control in community-dwelling older veterans with diabetes and comorbid dementia

Tightly controlled Poorly controlled HbA1c not (HbA1c ,7% [53 mmol/mol]) (HbA1c $9% [75 mmol/mol]) monitored OR 95% CI P value OR 95% CI P value OR 95% CI P value Male sex 1.10 0.83–1.46 0.49 1.02 0.59–1.75 0.94 0.81 0.45–1.46 0.48 Race/ethnicity White non-Hispanic (reference) dd ddd dddd Black non-Hispanic 0.91 0.81–1.02 0.11 1.68 1.39–2.04 ,0.001 1.16 0.91–1.49 0.24 Hispanic 0.77 0.67–0.87 ,0.001 1.58 1.28–1.96 ,0.001 0.96 0.72–1.28 0.79 Other 0.94 0.75–1.19 0.62 1.73 1.21–2.49 0.003 1.62 1.07–2.44 0.022 Age 65–74 years (reference) dd ddd dddd 75–84 years 1.16 1.07–1.26 0.001 0.81 0.70–0.94 0.005 1.29 1.06–1.56 0.011 $85 years 1.13 1.02–1.25 0.021 0.66 0.54–0.81 ,0.001 1.43 1.13–1.80 0.002 Has medication copay 0.97 0.90–1.04 0.34 0.99 0.86–1.13 0.83 0.84 0.72–0.99 0.036 Congestive heart failure 0.84 0.76–0.92 ,0.001 1.07 0.90–1.28 0.43 1.09 0.88–1.35 0.42 Heart valve disease 1.16 1.01–1.32 0.033 0.91 0.70–1.20 0.51 1.22 0.91–1.64 0.19 Peripheral vascular disease 1.03 0.94–1.13 0.56 0.86 0.72–1.03 0.11 0.84 0.67–1.05 0.13 Hypertension 0.96 0.88–1.05 0.38 1.08 0.91–1.29 0.37 0.39 0.33–0.45 ,0.001 Chronic lung disease 1.10 1.01–1.21 0.038 1.00 0.84–1.18 0.99 1.08 0.88–1.32 0.48 Hypothyroidism 1.03 0.92–1.15 0.62 0.96 0.77–1.19 0.68 0.60 0.44–0.82 0.001 Renal failure 0.90 0.82–1.00 0.045 1.13 0.94–1.35 0.21 0.78 0.61–0.99 0.045 Solid tumor without metastasis 0.99 0.90–1.09 0.85 0.90 0.74–1.10 0.31 0.67 0.52–0.87 0.003 Obesity 0.83 0.72–0.95 0.007 1.07 0.85–1.35 0.56 0.57 0.38–0.84 0.004 Weight loss 1.36 1.09–1.69 0.006 0.93 0.45–1.19 0.21 1.77 1.14–2.74 0.010 Fluid and electrolyte disorder 0.97 0.87–1.08 0.57 1.11 0.91–1.36 0.29 1.09 0.83–1.41 0.55 Deficiency anemia 1.12 1.02–1.22 0.016 0.76 0.64–0.92 0.003 0.78 0.63–0.98 0.030 Psychoses 1.08 0.97–1.20 0.14 1.04 0.86–1.26 0.68 0.87 0.68–1.11 0.26 Depression 1.05 0.95–1.15 0.35 1.05 0.88–1.25 0.60 0.91 0.72–1.15 0.45 Inpatient stay in FY2008 1.04 0.93–1.15 0.50 1.45 1.21–1.73 ,0.001 0.90 0.70–1.17 0.44 VA documentation of dementia diagnosis 1.05 0.98–1.14 0.18 0.99 0.86–1.16 0.94 0.97 0.82–1.15 0.71

Multinomial logistic regression was used, with moderate control (HbA1c $7% [53 mmol/mol] and ,9% [75 mmol/mol]) as the reference category. OR, odds ratio. care.diabetesjournals.org Thorpe and Associates 593

Table 3—Independent predictors of use of antidiabetic medications with high risk document the very high rate of use of of hypoglycemia in community-dwelling older veterans with diabetes and sulfonylureas and insulin in tightly con- comorbid dementia with tight glycemic control trolled patients with dementia, an im- OR 95% CI P value portant and modifiable contributor to ’ Male sex 1.77 1.19–2.63 0.004 these patients risk for hypoglycemia. We identified key risk factors for tight Race/ethnicity White non-Hispanic (reference) dd dglycemic control in patients with de- Black non-Hispanic 1.27 1.05–1.53 0.015 mentia. Age $75, weight loss, chronic Hispanic 0.92 0.75–1.13 0.42 lung disease, and deficiency anemias in- Other 0.84 0.60–1.16 0.28 crease the odds of tight glycemic con- Age trol, whereas obesity is protective. 65–74 years (reference) dd dProviders and patients may not recog- – – , 75 84 years 1.28 1.13 1.45 0.001 nize that changes in appetite and weight $85 years 1.60 1.37–1.88 ,0.001 associated with dementia itself, advanc- – Has medication copay 0.93 0.83 1.04 0.20 ing age, or other comorbidities may – , Congestive heart failure 1.51 1.28 1.78 0.001 make it easier to control blood glucose – Valvular disease 0.88 0.72 1.08 0.22 with less intense antidiabetic regimens Peripheral vascular disease 1.2 1.05–1.41 0.010 or that medication may no longer be re- Hypertension 1.10 0.97–1.26 0.14 quired. The results suggest that inter- Chronic lung disease 0.94 0.82–1.08 0.38 ventions should encourage the review Hypothyroidism 0.91 0.77–1.08 0.28 and deintensification of medication reg- Renal failure 4.60 3.67–5.78 ,0.001 imens in patients who lose weight or Solid tumor without metastasis 0.97 0.83–1.13 0.67 reach advanced ages. Obesity 1.01 0.82–1.26 0.91 We also found that patients with con- Weight loss 0.88 0.65–1.20 0.43 comitant congestive heart failure or re- Fluid and electrolyte disorder 0.83 0.70–1.00 0.047 nal failure exhibited lower odds of tight Deficiency anemia 0.96 0.84–1.10 0.59 control compared with dementia pa- Psychoses 0.94 0.81–1.10 0.47 tients without these comorbidities; Depression 0.85 0.74–0.98 0.026 thesediagnosesarenotedinVA/DoD Inpatient stay in FY2008 1.26 1.06–1.49 0.008 guidelines as indications to avoid tight VA documentation of dementia 1.09 0.97–1.22 0.16 glycemic control. The explicit mention of the role of dementia in setting glyce- Tight glycemic control, HbA ,7% (53 mmol/mol). OR, odds ratio. 1c mic control targets, as in the newly re- vised 2013 AGS guidelines (20) and a recently proposed quality indicator for CONCLUSIONS The study has several important diabetes overtreatment (22), may be an In a national cohort of .15,000 outpa- strengths. Despite growing numbers of essential first step in encouraging ap- tients with medication-treated type 2 older diabetic patients with dementia propriate diabetes treatment deintensi- diabetes and dementia, we found that and their heightened vulnerability to fication in this population. Although the more than one-half had HbA1c ,7% (53 tight control, few studies have exam- present findings suggest some contrib- mmol/mol), despite clear guidelines rec- ined glycemic control levels partly be- uting factors to the lack of appropriate ommending higher glycemic targets. In cause large enough data sources deintensification of diabetes treatment addition, 75% of tightly controlled pa- containing HbA1c values for this popula- in patients with dementia, future re- tients with dementia used medications tion are rare outside the VA. A recent search should directly engage providers, that further exacerbated their potential studydocumentedhighprevalence patients, and caregivers to uncover their risk for hypoglycemia. Even in the con- (45–50%) of tight glycemic control gen- perceptions of how dementia should al- text of heightened hypoglycemia risk erally among veterans with a range of ter glycemic control targets and barriers resulting from the combination of co- risk factors for hypoglycemia, including to pursuing less intensive targets as well morbid dementia and tight glycemic dementia (31). The present study ex- as what role guidelines play in these control, the results suggest that the tends this work by focusing specifically decisions. large majority of providers do not sub- on dementia status and examining risk The present findings also highlight the stitute sulfonylureas and insulin with factors for tight control or sulfonylurea/ need for initiatives to support safer an- lower-risk antidiabetic agents. These insulin use and by using Medicare data tidiabetic prescribing choices for pa- findings highlight the need for inter- to help to identify patients with dementia. tients with dementia. The widespread ventions to encourage appropriate The incorporation of Medicare claims in use of sulfonylureas and insulin likely deintensificationandalterationof identifying dementia patients is critical given reflects VA/DoD guidelines recommend- medications when diabetic patients past research demonstrating its importance ing both options as first-line antidiabetic develop dementia, which has not in accurately capturing comorbidities for therapies along with metformin and as been reflectedinpreviousdiabetes older veterans (32). In addition, we be- add-on therapies to metformin or each quality initiatives. lieve that the present study is the first to other if glycemic goals are not achieved, 594 Intense Diabetes Treatment in Dementia Patients Diabetes Care Volume 38, April 2015

with no discussion of how dementia higher HbA1c targets for older patients patients may have improved since, espe- should affect medication choice. The with dementia, some patients with mild cially in light of increased attention to the greatly increased odds of sulfonylureas dementia in otherwise excellent health possible risks of tight control and hypogly- and/or insulin use in patients with con- and good functional status may opt for cemia in older patients with comorbidities. gestive heart failure and renal failure tighter glycemic targets. Although we Finally, the results reflect a primarily male and those aged $75 years also likely did capture extensive information on veteran population and may not general- reflect VA/DoD (and other) guidelines, the presence of comorbid conditions, ize to women or nonveterans. which list these conditions and age we were not able to account for patients’ In conclusion, the results show a high $80 years as contraindications to the functional status or preferences in the rate of intense treatment of diabetes in use of metformin, leaving sulfonylurea analyses. Likewise, insulin therapy may older patients with comorbid dementia, and insulin as the preferred agents over be appropriate for older dementia pa- potentially placing them at elevated risk other classes of agents with lower hypo- tients with long-standing diabetes who for hypoglycemia and serious adverse glycemic risk. Although these other are unable to attain HbA1c targets with events. Equally disconcerting is the high antidiabetic medication classes (i.e., noninsulin agents alone. We did not ex- frequency of use of medications known TZDs, DPP-4 inhibitors, GLP-1 agonists, amine dose of medications; thus, it is to cause hypoglycemia. The findings a-glucosidase inhibitors) have lower possible that some tightly controlled pa- present a compelling need for the devel- associated risk of hypoglycemia, they tients were prescribed very low doses. opment of quality initiatives to encourage have lower efficacy in reducing HbA1c However, by limiting the analysis of hy- review of glycemic targets and antidia- and lack robust evidence regarding their poglycemic medication use to patients betic medications in this rapidly growing safety profile and associated treatment with HbA1c ,7% (53 mmol/mol), such group of patients, especially those aged burden in older patients with dementia. patients face a particularly high risk of $75 years and those with weight loss. Although these alternative agents were hypoglycemia and may benefit from de- not on the VA national formulary at the intensification of insulin dose or a switch time of the study (except for acarbose), to a non–insulin-containing regimen. Funding. C.T.T., S.Z., M.M., and M.J.F. are they were available to all prescribers Although this study is unique in cap- supported by VA Health Services Research & De- when necessary through a nonformu- turing such a large number of patients velopment (CIN 13-405). W.F.G. is supported by a lary request process. Because clinical with both diabetes and dementia as well VA Health Services Research & Development Ca- trials comparing various antidiabe- as important clinical variables like reer Development Award (CDA 09-207). tic medications in older patients with HbA , we acknowledge that the obser- The views expressed in this article are those of 1c the authors and do not necessarily represent the dementia are unlikely to occur, well- vational study design, relying solely on views of the Department of Veterans Affairs. designed observational studies are needed administrative data, has inherent limita- Duality of Interest. No potential conflicts of to guide prescribing choices for this rapidly tions. In particular, we were not able to interest relevant to this article were reported. growing population. In the meantime, the capture the full range of covariates that Author Contributions. C.T.T. contributed to present results suggest a need for provider may be associated with glycemic control the study concept and design, study supervi- sion, data acquisition, data analysis and in- outreach efforts to consider medication and medication choice, such as assistance terpretation, statistical analysis, drafting of the options with a lower hypoglycemic risk with at-home diabetes self-management manuscript, and critical revision of the manu- and encourage the use of the minimally from informal family caregivers or nurse- script for important intellectual content. W.F.G. intensive regimen required to achieve provided home care. We also did not ex- and X.Z. contributed to the study concept and design, data acquisition, data analysis and in- moderate glycemic control levels. amine interactions between covariates terpretation, and critical revision of the manu- The results of this study should be because of a lack of theory to guide script for important intellectual content. C.B.G. interpreted in light of several limita- such analyses and relatively small num- contributed to the study concept and design, tions. First, our focus was on under- bers of patients with some comorbid con- data analysis and interpretation, and critical standing patient characteristics associated ditions (e.g., weight loss). Future research revision of the manuscript for important in- tellectual content. S.Z. contributed to the data withtwoestablishedriskfactorsforhypo- may consider possible multiplicative ef- acquisition, data analysis and interpretation, glycemia (i.e., HbA1c ,7% [53 mmol/mol], fects of specific comorbid conditions or and critical revision of the manuscript for use of sulfonylureas and/or insulin), and other risk factors identified in this study. important intellectual content. M.M. contrib- uted to the study supervision, data acquisition, we did not capture data on actual hypogly- Patients may also have had HbA1c tests cemic events. Whether deintensification of and medications filled outside the VA data analysis and interpretation, and critical revision of the manuscript for important in- therapy in response to tight glycemic con- that we were unable to capture, although tellectual content. M.J.F. contributed to the trol or substitution of insulin and sulfonyl- we reduced this possibility by limiting the study supervision; administrative, technical, urea with other antidiabetic agents results sample to patients who filled antidiabetic and material support; data analysis and in- in reduced hypoglycemic events in diabetic medications in the VA and did not have terpretation; and critical revision of the manu- patients with dementia is an important di- another major source of drug coverage. In script for important intellectual content. C.T.T. is the guarantor of this work, and as such, had rection for future research. In addition, the addition, we used a single HbA1c value to full access to all the data in the study and takes relationship between glycemic control lev- classify patients’ glycemic control at one responsibility for the integrity of the data and els and the progression of dementia is un- point in time and therefore cannot deter- the accuracy of the data analysis. known. Whether tight glycemic control mine the extent to which their glycemic Prior Presentation. Parts of this study were presented in poster form at the 2013 Annual hinders or hastens dementia progression control levels were transient versus stable. Scientific Meeting of the Gerontological Society should be explored in future studies. We also acknowledge that the data are of America, New Orleans, LA, 20–24 November Second, although guidelines recommend several years old, and care of dementia 2013. care.diabetesjournals.org Thorpe and Associates 595

References 12. Hewitt J, Smeeth L, Chaturvedi N, Bulpitt CJ, medications in the VA healthcare system. Arch In- 1. Boyle JP, Honeycutt AA, Narayan KMV, et al. Fletcher AE. Self management and patient un- tern Med 2012;172:1608–1609 Projection of diabetes burden through 2050: im- derstanding of diabetes in the older person. Di- 24. Gellad WF, Donohue JM, Zhao X, et al. pact of changing demography and disease preva- abet Med 2011;28:117–122 Brand-name prescription drug use among Vet- lence in the U.S. Diabetes Care 2001;24:1936–1940 13. Briscoe VJ, Davis SN. Hypoglycemia in type 1 erans Affairs and Medicare Part D patients with 2. Feil DG, Rajan M, Soroka O, Tseng CL, Miller and type 2 diabetes: physiology, pathophysiology, diabetes: a national cohort comparison. Ann In- DR, Pogach LM. Risk of hypoglycemia in older and management. Clin Diabetes 2006;24:115–121 tern Med 2013;159:105–114 veterans with dementia and cognitive impair- 14. Xie J, Brayne C, Matthews FE; Medical Re- 25. Miller DR, Safford MM, Pogach LM. Who ment: implications for practice and policy. J search Council Cognitive Function and Ageing Study has diabetes? Best estimates of diabetes prev- Am Geriatr Soc 2011;59:2263–2272 Collaborators. Survival times in people with demen- alence in the Department of Veterans Affairs 3. Thorpe CT, Thorpe JM, Kind AJ, Bartels CM, tia: analysis from population based cohort study based on computerized patient data. Diabetes Everett CM, Smith MA. Receipt of monitoring of with 14 year follow-up. BMJ 2008;336:258–262 Care 2004;27(Suppl. 2):B10–B21 diabetes mellitus in older adults with comorbid 15. Williams MM, Xiong C, Morris JC, Galvin JE. 26. Buccaneer Computer Systems and Service, dementia. J Am Geriatr Soc 2012;60:644–651 Survival and mortality differences between de- Inc. Chronic Conditions Data Warehouse user 4. Yaffe K, Falvey CM, Hamilton N, et al.; Health mentia with Lewy bodies vs Alzheimer disease. manual [Internet]. Version 3.0, 2014. Available ABC Study. Association between hypoglycemia 2006;67:1935–1941 from https://www.google.com/url?sa5t& and dementia in a biracial cohort of older adults 16. Rait G, Walters K, Bottomley C, Petersen I, rct5j&q5&esrc5s&source5web&cd53&ved5 with diabetes mellitus. JAMA Intern Med 2013; Iliffe S, Nazareth I. Survival of people with clin- 0CCwQFjAC&url5https%3A%2F%2Fwww 173:1300–1306 ical diagnosis of dementia in primary care: co- .ccwdata.org%2Fcs%2Fgroups%2Fpublic% 5. Yaffe K, Falvey C, Hamilton N, et al. Diabetes, hort study. BMJ 2010;341:c3584 2Fdocuments%2Fdocument%2Fccw_userguide glucose control, and 9-year cognitive decline 17. Department of Veterans Affairs; Depart- .pdf&ei5QmihVMSOCoXQggTN1YH4Dg& among older adults without dementia. Arch ment of Defense. VA/DoD Clinical Practice usg5AFQjCNHhSApHILlPDoY1vvlrCP-Ckbxhzw& Neurol 2012;69:1170–1175 Guideline: Management of Diabetes Mellitus sig25Lbk_uTd_AfN-P1yWz-EqDA&bvm5 6. Biessels GJ, Staekenborg S, Brunner E, (DM). Version 4.0. Washington, DC, Veterans bv.82001339,d.eXY&cad5rja. Accessed 29 Brayne C, Scheltens P. Risk of dementia in di- Health Administration, 2010 December 2014 abetes mellitus: a systematic review. Lancet 18. Department of Veterans Affairs; Depart- 27.TaylorDHJr.,ØstbyeT,LangaKM,WeirD, Neurol 2006;5:64–74 ment of Defense. VA/DoD Clinical Practice Plassman BL. The accuracy of Medicare claims 7. National Plan to Address Alzheimer’sDisease. Guideline: Management of Diabetes Mellitus as an epidemiological tool: the case of demen- Washington, DC, U.S. Department of Health and in Primary Care. Version 3.0. Washington, DC, tia revisited. J Alzheimers Dis 2009;17:807– Human Services, 2013 Veterans Health Administration, 2003 815 8. Bynum JP, Rabins PV, Weller W, Niefeld M, 19. American Diabetes Association. Standards 28. Eicheldinger C, Bonito A. More accurate Anderson GF, Wu AW. The relationship of medical care in diabetesd2014. Diabetes racial and ethnic codes for Medicare adminis- between a dementia diagnosis, chronic illness, Care 2014;37(Suppl. 1):S14–S80 trative data. Health Care Financ Rev 2008;29: Medicare expenditures, and hospital use. J Am 20. Moreno G, Mangione CM, Kimbro L, Vaisberg 27–42 Geriatr Soc 2004;52:187–194 E; American Society Expert Panel on 29. Elixhauser A, Steiner C, Harris DR, Coffey 9. Onder G, Landi F, Fusco D, et al. Recommen- Care of Older Adults with Diabetes Mellitus. RM. Comorbidity measures for use with admin- dations to prescribe in complex older adults: Guidelines abstracted from the American Geriat- istrative data. Med Care 1998;36:8–27 results of the CRIteria to assess appropriate rics Society Guidelines for Improving the Care of 30. Klabunde CN, Potosky AL, Legler JM, Medication use among Elderly complex patients Older Adults with Diabetes Mellitus: 2013 update. Warren JL. Development of a comorbidity index (CRIME) project. Drugs Aging 2014;31:33–45 J Am Geriatr Soc 2013;61:2020–2026 using claims data. J Clin Epidemiol 10. Feil DG, Zhu CW, Sultzer DL. The relation- 21. American Geriatrics Society. Choosing Wisely: 2000;53:1258–1267 ship between cognitive impairment and diabe- Five Things and Patients Should Ques- 31. Tseng CL, Soroka O, Maney M, Aron DC, tes self-management in a population-based tion. New York, American Geriatrics Society, 2013 Pogach LM. Assessing potential glycemic over- community sample of older adults with type 2 22. Pogach L, Aron D. The other side of quality treatment in persons at hypoglycemic risk. diabetes. J Behav Med 2012;35:190–199 improvement in diabetes for seniors: a proposal JAMA Intern Med 2014;174:259–268 11. Kirsh SR, Aron DC. Choosing targets for gly- for an overtreatment glycemic measure. Arch 32. Byrne MM, Kuebeler M, Pietz K, Petersen caemia, blood pressure and low-density lipopro- Intern Med 2012;172:1510–1512 LA. Effect of using information from only one tein cholesterol in elderly individuals with 23. Gellad W, Mor M, Zhao X, Donohue J, Good C. system for dually eligible health care users. diabetes mellitus. Drugs Aging 2011;28:945–960 Variation in use of high-cost diabetes mellitus Med Care 2006;44:768–773